AI for real work · 3 of 5 · Prepare
Is your business ready for AI? What to review before you start
After choosing an opportunity, you want to see it work. You can already picture an improvement and want to move towards it. At this point, a brief pause helps us understand what the idea needs to become something people can actually use.
Preparation addresses that distance. A technical capability may be available while the business still needs clearer information, an unresolved decision or time from people who understand the work.
Recognising those conditions helps us organise the next steps. Every requirement understood early improves our ability to decide what to build and how to build it.
Be ready for a specific improvement
The question of readiness becomes useful when we connect it to the chosen project. Its scope determines the information we need, the consequences of an error and the people who should participate.
We therefore begin by revisiting what we want to achieve and describing the conditions in which the solution would be used. A better understanding of that context helps distinguish what is essential to start from what can be developed later.
The assessment should also consider what we will be able to verify. If we cannot yet recognise an acceptable result, we need to establish that criterion before judging the performance of AI.
This is a first form of preparation: reaching sufficient agreement about the work we want to improve. That agreement guides what follows and allows people to evaluate the same proposal.
Make information a usable resource
A business can hold a great deal of information and still struggle to use it. Its value depends on being accessible, understandable and appropriate for the task.
Before incorporating it into an AI application, we should identify which sources are authoritative, what their data means and under which conditions it remains valid. We also need to know who can resolve a contradiction or fill a gap.
This work requires business judgement. A well-written answer may rely on a source that is no longer applicable. Correct information may have a limited scope that is lost when it is used in a different context. Preparing sources includes preserving these relationships.
People’s knowledge is also part of the available resources. When something important depends on experience that has not been documented, we need a way to make it explicit and review it. AI can assist with that process, while people who understand the work provide the judgement needed to validate the record.
Information then begins to acquire the conditions needed to support a solution: a recognisable origin, a clear meaning and someone who can maintain it.
Understand the process that will change
Information is used within a sequence of work. Integrating it with AI requires understanding how that sequence begins, which transformations it involves and what result it must deliver.
We also need to examine the decisions people make today. Some follow clear rules; others require interpreting the context or consulting someone. Making these differences visible helps us decide what the solution can assist with and what review should remain.
As we define that division of responsibilities, exceptions emerge. We need to understand what happens when information is missing, a request exceeds the scope or a result raises doubts. The process needs a way to continue under those conditions.
Its connections matter too. If the work uses other systems, information must move between stages with the appropriate meaning and format. Preparing an integration starts by understanding what each part needs to receive.
Make room for people
An improvement changes someone’s work in some way. That person needs to understand what it is for, how to use it and what responsibility they retain for the result.
Their involvement from the beginning reveals requirements that a general description may miss. It also helps us assess whether the proposal reduces a difficulty or shifts it to another part of the process.
That participation takes time, which needs to be reserved. People who know the work must be able to explain, review and test. Those who will use the solution need to practise checking it and resolving uncertainties. Preparation includes making room for that learning within the activity.
We must also agree who makes decisions about the process, who maintains the information and who handles technical problems. One person may take on several responsibilities, but each responsibility needs a clear place. The ability to continue after the first test depends on this.
Give trust a basis we can verify
Trusting a solution requires understanding the conditions under which we can use it. We need controls proportionate to the work and the consequences of an error.
We should define what information each person may access, how it will be processed and what records are needed to review a result. We also need a way to ask for help and continue working if the assistance is unavailable.
The NIST AI Risk Management Framework considers context and risk throughout a system’s life cycle. For a particular initiative, that approach can translate into understandable decisions about access, review, responsibilities and monitoring.
Trust develops when those decisions can be checked in practice. We need to know who will review, what they will be able to observe and what they will do when something falls outside the expected conditions.
Turn what is missing into a plan
The review will leave us with some conditions resolved and others outstanding. A useful next step is to state what is missing, why it matters, who can resolve it and what evidence will establish that it is complete.
Some requirements must be met before processing information or involving users. Others can be investigated within a bounded test. Distinguishing them lets us proceed with a scope the business can support.
Preparation fulfils its purpose when a general impression becomes a set of decisions people can act on. With that foundation, we can design a test that produces useful evidence.